Research

I work on problems that bridge theoretical foundations with practical systems, with a focus on AI and quantum computing.

AI Research

Mechanistic Interpretability

Interpretability / Safety

Developing methods to understand what neural networks learn. My sensitivity-based feature discovery work proposes alternatives to sparse autoencoders that explain feature sparsity without requiring it as an assumption.

Formal Proofs for Transformers

ML Theory

Proving formal statements about transformer behavior given their weights. Leveraging techniques from verification and formal methods to make rigorous claims about specific neural network computations.


Quantum Cryptography & Computing

One-Time Programs & Quantum Cryptography

Quantum Crypto

Building cryptographic primitives from physics rather than computational assumptions. Recent work includes verifiable one-time programs enabling single-round secure computation (arXiv), constructions without long-term quantum memory (arXiv), information-theoretic security from geometrically local adversaries (arXiv), and one-time memories from stateless hardware (arXiv).

Depth Bounded Quantum Cryptography

Quantum Crypto

Exploring how quantum cryptography becomes more powerful when adversaries have limited quantum resources — a realistic model for near-term security.

Quantum Error Correction

Quantum Computing

Work on weight-reduced codes, hypergraph products, and improved decoding algorithms. Includes effective distance preservation for weight reduced codes (arXiv) and improved Small-Set-Flip decoding presented at QIP 2024 (arXiv).


Selected Past Work

Multiparty Secret Leader Election

Cryptography

Work done while at the Ethereum Foundation with Mark Simkin to improve communication cost of secret leader election for multiple leaders. Resulted in a novel data independent and oblivious priority queue.

Weighted Secret Sharing from the Wiretap Channel

Information Theory

With Fabrice Benhamouda and Shai Halevi, introduced a novel connection between weighted secret sharing and wiretap channels. Presented at Information-Theoretic Cryptography (ITC) 2023 (ePrint).

3D Turtles and Drawing Math

Math Visualization

Explored whether movement sequences generated by rational numbers form closed paths. The closure question connects to discrete logs and specific multinomial roots. Paper | Interactive demo

3D Turtle visualization